Summary:

ShadowView's constructor reads the node's layout metrics through layoutMetricsFromShadowNode, which uses dynamic_cast<const LayoutableShadowNode *> to find out whether the node has layout. The differ builds a ShadowView for every child it visits when it slices a node's children, so this runs for every visited node on every commit, including the per-frame commits that update a ScrollView's content offset.

Current behavior. Every call pays for a dynamic_cast, which walks the type's inheritance information at runtime; for a View, the node's type is ConcreteViewShadowNode<...>, several levels below LayoutableShadowNode. In Time Profiler traces of scrolling, __dynamic_cast took 19–21% of the differ's time.

With this change. ShadowNode gets a virtual asLayoutable() that returns nullptr, and LayoutableShadowNode overrides it to return this. layoutMetricsFromShadowNode calls it instead of the dynamic_cast, so every node, not only Views, gets its layout metrics without a runtime type check, and the compiler guarantees the result. LayoutableShadowNode::getDebugProps is marked override, since clang requires it once the class has another override.

Changelog:

[GENERAL] [CHANGED] - Building a ShadowView no longer uses a dynamic_cast to read a node's layout metrics

Test Plan:

• The pinned clang-format passes on the changed files. The C++ API snapshots in scripts/cxx-api/api-snapshots list the new method.


• 
  yarn fantom (full suite, with the tester rebuilt from this change): 3,289 tests pass (110 skipped by their configuration). In that run the tester crashed (SIGSEGV) in two other files, AnimatedBackend-itest and StaleEventHandlersFromInterruptedRender-itest; neither crashed in 30 runs each on their own, and the same kind of crash happens locally without this change.


• 
  Measurement. RNTester built with xcodebuild in Release and run on a physical iPhone 15 Pro (iOS 26.6.1). The two arms are the same binary with the change switched by a launch argument, in the order before, after, after, before. Interactions were driven by Argent on the physical iPhone. Timed: the diff phase of each mount transaction from TransactionTelemetry, on the main thread.

RNTester Playground
before
this PR
change

A FlatList of 500 multi-line messages, 20 flings down and 20 up per launch (~8,700 transactions): diff time per launch
1,140 ms
1,006 ms
−11.8%

Same, diff time per 1–2-mutation transaction (the per-frame scroll updates)
111.0 µs
98.3 µs
−11.4%

1000 labels mounted and unmounted, 10 times per arm: diff time per mount or unmount (median)
2.61 ms
2.47 ms
−5.4%

🤖 Generated with Claude Code